Proposal on building tensorboard-like tool

[VoVAllen]

By using Kibana+Logstash tech stack. It's a very common and matured tool(also open-sourced) in industry. It provides the interface to show&collect logs from various machines. Therefore what we only need to do is design the format of mxnet's output log about what to export to the tool.

And it can be deployed easily&independently by docker.

Just a very naive idea. But I believe this tool will makes mxnet easier to use. It's also useful in real industry community.

[piiswrong]

logging and visualization is something we should have. I don't know about Kibana+Logstash though. Seems like pretty heavy machinery with a log of dependencies. Is it really necessary?

[VoVAllen]

Yes, you are right……They are much bigger than I thought(It needs about 150MB, most part may be the jvm in docker......). I'll keep on finding appropriate tools to do so.

[VoVAllen]

However I would say logging and visualization are kind of independent. Maybe we could start from logging first.

[piiswrong]

yes. We can even consider using tensorflow's logging format and try to strip tensorboard out of tensorflow

[austingg]

@piiswrong for visualizing log, and ploting training curves, you may reference nvidia-digits, there is a light-weight web-based module based on python flask.

[yajiedesign]

This issue is closed due to lack of activity in the last 90 days. Feel free to reopen if this is still an active issue. Thanks!